Cluster Reconstruction in Electromagnetic Calorimeters Using Machine Learning Methods

Abstract

Machine-learning-based methods can be developed for the reconstruction of clusters in segmented detectors for high energy physics experiments. Convolutional neural networks with autoencoder architecture trained on labeled data from a simulated dataset reconstruct events by providing information about the hit point and energy of each particle that has entered the detector. The correct reconstruction of the positionand the energy of the incident particles is crucial for the accurate events reconstruction. The presented method shows the ability to reconstruct the impact point within the same segment as the true position and determines the particle energy with good precision. It can be applied in a wide range of cases of event reconstruction where the good separation of overlapping signals plays a key role in the data analysis.